03.10.2016 3:47 PM "Henri Verbeet" hverbeet@gmail.com:
I think it may be harder to fix than this. If state_unbind_resources() can end up destroying the (primary) swapchain, it's not safe to call context_acquire() afterwards. Which means it's unsafe to unload resources or even unload a different SRV. The reason destroying the RTVs is currently safe, is that we unloaded resources earlier, so releasing the references to the resources for those views shouldn't acquire a context to destroy e.g. the GL texture again, but that's already more fragile than I'd like. Perhaps we should just delay destroying the context until leaving the last level.
I think this fix is correct. The device keeps a reference to the implicit swapchain. The implicit swapchain is destroyed just before leaving from wined3d_device_uninit_3d() so it should be safe to call context_acquire().
This problem occurs only for additional swapchains and that's why it happens only in test_swapchain_flip().